Play Style Identification Using Low-Level Representations of Play Traces in MicroRTS

2025-07-14

AI Summary

This work addresses the over-reliance on hand-crafted features and domain knowledge in gameplay style identification for game AI. We propose an unsupervised representation learning method tailored to low-level action trajectories in MicroRTS. Specifically, we design a CNN-LSTM autoencoder that directly learns compact, semantically rich latent representations from raw action sequences—bypassing manual feature engineering and explicit domain abstractions. In a fully unsupervised setting (i.e., without labels), our approach achieves separable clustering of diverse AI strategies in the learned latent space. Furthermore, we demonstrate that this space effectively supports both style discrimination and controllable generation of behaviorally diverse policies. Experiments confirm that the framework substantially reduces dependence on domain priors, offering a novel pathway toward interpretable game AI behavioral modeling and adaptive content generation.

Technology Category

• Humans and AI: Game Design — Procedural Content Generation & Storytelling
• Machine Learning: Representation Learning
• Game Theory and Economic Paradigms: Adversarial Learning

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Search and Retrieval-Augmented AI: Web query analysis, representation and understanding

Abstract

Play style identification can provide valuable game design insights and enable adaptive experiences, with the potential to improve game playing agents. Previous work relies on domain knowledge to construct play trace representations using handcrafted features. More recent approaches incorporate the sequential structure of play traces but still require some level of domain abstraction. In this study, we explore the use of unsupervised CNN-LSTM autoencoder models to obtain latent representations directly from low-level play trace data in MicroRTS. We demonstrate that this approach yields a meaningful separation of different game playing agents in the latent space, reducing reliance on domain expertise and its associated biases. This latent space is then used to guide the exploration of diverse play styles within studied AI players.

Problem

• Identifies play styles from low-level game traces
• Reduces reliance on domain-specific handcrafted features
• Enables exploration of diverse AI player behaviors

Innovation

• Unsupervised CNN-LSTM autoencoder for latent representations
• Direct use of low-level play trace data
• Latent space guides diverse play style exploration